[dm-thin] Fix a race condition between one thin device discarding a b…

…lock and another provisioning it.

The discard passdown was being issued after the unmapping, which meant
the block could be reprovisioned whilst the passdown discards are
still in flight.

Now the unmapping does not occur until after the passdown is complete.

drivers/md/dm-thin.c

@@ -253,6 +253,7 @@ struct pool {
 	struct bio_list deferred_flush_bios;
 	struct list_head prepared_mappings;
 	struct list_head prepared_discards;
+	struct list_head prepared_discards_pt2;
 	struct list_head active_thins;
 
 	struct dm_deferred_set *shared_read_ds;
@@ -269,6 +270,7 @@ struct pool {
 
 	process_mapping_fn process_prepared_mapping;
 	process_mapping_fn process_prepared_discard;
+	process_mapping_fn process_prepared_discard_pt2;
 
 	struct dm_bio_prison_cell **cell_sort_array;
 };
@@ -1014,7 +1016,8 @@ static void process_prepared_discard_no_passdown(struct dm_thin_new_mapping *m)
 
 /*----------------------------------------------------------------*/
 
-static void passdown_double_checking_shared_status(struct dm_thin_new_mapping *m)
+static void passdown_double_checking_shared_status(struct dm_thin_new_mapping *m,
+						   struct bio *discard_parent)
 {
 	/*
 	 * We've already unmapped this range of blocks, but before we
@@ -1027,7 +1030,7 @@ static void passdown_double_checking_shared_status(struct dm_thin_new_mapping *m
 	dm_block_t b = m->data_block, e, end = m->data_block + m->virt_end - m->virt_begin;
 	struct discard_op op;
 
-	begin_discard(&op, tc, m->bio);
+	begin_discard(&op, tc, discard_parent);
 	while (b != end) {
 		/* find start of unmapped run */
 		for (; b < end; b++) {
@@ -1062,7 +1065,61 @@ static void passdown_double_checking_shared_status(struct dm_thin_new_mapping *m
 	end_discard(&op, r);
 }
 
-static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m)
+static void queue_passdown_pt2(struct dm_thin_new_mapping *m)
+{
+	unsigned long flags;
+	struct pool *pool = m->tc->pool;
+
+	spin_lock_irqsave(&pool->lock, flags);
+	list_add_tail(&m->list, &pool->prepared_discards_pt2);
+	spin_unlock_irqrestore(&pool->lock, flags);
+	wake_worker(pool);
+}
+
+static void passdown_endio(struct bio *bio)
+{
+	/*
+	 * It doesn't matter if the passdown discard failed, we still want
+	 * to unmap (we ignore err).
+	 */
+	queue_passdown_pt2(bio->bi_private);
+}
+
+static void process_prepared_discard_passdown_pt1(struct dm_thin_new_mapping *m)
+{
+	int r;
+	struct bio *discard_parent;
+	struct thin_c *tc = m->tc;
+
+	/*
+	 * We're about to do passdown, so we should consider the original
+	 * bio to be in flight.
+	 */
+	inc_all_io_entry(tc->pool, m->bio);
+
+	discard_parent = bio_alloc(GFP_NOIO, 1);
+	if (!discard_parent) {
+		DMWARN("unable to allocate top level discard bio for passdown.  Skipping passdown.");
+		queue_passdown_pt2(m);
+
+	} else {
+		discard_parent->bi_end_io = passdown_endio;
+		discard_parent->bi_private = m;
+
+		if (m->maybe_shared)
+			passdown_double_checking_shared_status(m, discard_parent);
+
+		else {
+			struct discard_op op;
+			begin_discard(&op, tc, discard_parent);
+			r = issue_discard(&op, m->data_block,
+					  m->data_block + (m->virt_end - m->virt_begin));
+			end_discard(&op, r);
+		}
+	}
+}
+
+static void process_prepared_discard_passdown_pt2(struct dm_thin_new_mapping *m)
 {
 	int r;
 	struct thin_c *tc = m->tc;
@@ -1073,16 +1130,8 @@ static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m)
 		metadata_operation_failed(pool, "dm_thin_remove_range", r);
 		bio_io_error(m->bio);
 
-	} else if (m->maybe_shared) {
-		passdown_double_checking_shared_status(m);
-
-	} else {
-		struct discard_op op;
-		begin_discard(&op, tc, m->bio);
-		r = issue_discard(&op, m->data_block,
-				  m->data_block + (m->virt_end - m->virt_begin));
-		end_discard(&op, r);
-	}
+	} else
+		bio_complete(m->bio, r);
 
 	cell_defer_no_holder(tc, m->cell);
 	mempool_free(m, pool->mapping_pool);
@@ -2227,6 +2276,8 @@ static void do_worker(struct work_struct *ws)
 	throttle_work_update(&pool->throttle);
 	process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard);
 	throttle_work_update(&pool->throttle);
+	process_prepared(pool, &pool->prepared_discards_pt2, &pool->process_prepared_discard_pt2);
+	throttle_work_update(&pool->throttle);
 	process_deferred_bios(pool);
 	throttle_work_complete(&pool->throttle);
 }
@@ -2355,7 +2406,8 @@ static void set_discard_callbacks(struct pool *pool)
 
 	if (passdown_enabled(pt)) {
 		pool->process_discard_cell = process_discard_cell_passdown;
-		pool->process_prepared_discard = process_prepared_discard_passdown;
+		pool->process_prepared_discard = process_prepared_discard_passdown_pt1;
+		pool->process_prepared_discard_pt2 = process_prepared_discard_passdown_pt2;
 	} else {
 		pool->process_discard_cell = process_discard_cell_no_passdown;
 		pool->process_prepared_discard = process_prepared_discard_no_passdown;
@@ -2840,6 +2892,7 @@ static struct pool *pool_create(struct mapped_device *pool_md,
 	bio_list_init(&pool->deferred_flush_bios);
 	INIT_LIST_HEAD(&pool->prepared_mappings);
 	INIT_LIST_HEAD(&pool->prepared_discards);
+	INIT_LIST_HEAD(&pool->prepared_discards_pt2);
 	INIT_LIST_HEAD(&pool->active_thins);
 	pool->low_water_triggered = false;
 	pool->suspended = true;